123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490 |
- # @Author : lightXu
- # @File : choice_box.py
- # @Time : 2018/11/22 0022 下午 16:01
- import re
- import time
- import xml.etree.cElementTree as ET
- import cv2
- import numpy as np
- from segment.sheet_resolve.analysis.choice.choice_m_row_column import get_choice_m_row_and_col
- from segment.sheet_resolve.tools import utils
- from segment.sheet_resolve.tools.brain_api import get_ocr_text_and_coordinate
- def get_interval(word_result_list):
- all_char_str = ''
- location = []
- for i, chars_dict in enumerate(word_result_list):
- chars_list = chars_dict['chars']
- for ele in chars_list:
- all_char_str = all_char_str + ele['char']
- location.append(ele['location'])
- pattern1 = re.compile(r"\]\[")
- pattern2 = re.compile(r"\[[ABCD]")
- def intervel(pattern):
- group_list = []
- for i in pattern.finditer(all_char_str):
- # print(i.group() + str(i.span()))
- group_list.append(list(i.span()))
- # print(group_list)
- sum_intervel = 0
- size = 0
- for group in group_list:
- left_x, right_x = location[group[0]]['left'] \
- + location[group[0]]['width'], location[group[1] - 1]['left']
- if abs(location[group[0]]['top'] - location[group[1]]['top']) < location[group[0]]['height']:
- if right_x - left_x > 0:
- sum_intervel = sum_intervel + right_x - left_x
- size += 1
- # print(sum_intervel // size)
- return sum_intervel // size
- intervel_width1 = intervel(pattern1)
- intervel_width2 = intervel(pattern2)
- return (intervel_width1 + intervel_width2) * 2 // 3
- def preprocess(image0, xe, ye):
- scale = 0
- dilate = 1
- blur = 5
- # 预处理图像
- img = image0
- # rescale the image
- if scale != 0:
- img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)
- # Convert to gray
- img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
- # # Apply dilation and erosion to remove some noise
- # if dilate != 0:
- # kernel = np.ones((dilate, dilate), np.uint8)
- # img = cv2.dilate(img, kernel, iterations=1)
- # img = cv2.erode(img, kernel, iterations=1)
- # Apply blur to smooth out the edges
- # if blur != 0:
- # img = cv2.GaussianBlur(img, (blur, blur), 0)
- # Apply threshold to get image with only b&w (binarization)
- img = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
- # cv2.namedWindow('image', cv2.WINDOW_NORMAL)
- # cv2.imshow('image', img)
- # if cv2.waitKey(0) == 27:
- # cv2.destroyAllWindows()
- # cv2.imwrite('otsu.jpg', img)
- kernel = np.ones((ye, xe), np.uint8) # y轴膨胀, x轴膨胀
- dst = cv2.dilate(img, kernel, iterations=1)
- # cv2.imshow('dilate', dst)
- # if cv2.waitKey(0) == 27:
- # cv2.destroyAllWindows()
- return dst
- def contours(image):
- _, cnts, hierarchy = cv2.findContours(image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
- bboxes = []
- for cnt_id, cnt in enumerate(reversed(cnts)):
- x, y, w, h = cv2.boundingRect(cnt)
- bboxes.append((x, y, x + w, y + h))
- return bboxes
- def box_coordinates(img):
- img_arr = np.asarray(img)
- def axix_break_point(img, tolerance_number, axis):
- sum_x_axis = img.sum(axis=axis)
- sum_x_axis[sum_x_axis > 255 * tolerance_number] = 1 # 白色有字
- sum_x_axis[sum_x_axis != 1] = 0 # 黑色无字
- sum_x_axis_list = list(sum_x_axis)
- sum_x_axis_list.append(0) # 最后几行到结束有字时,使索引值增加最后一位
- split_x_index = []
- num = 1
- for index, ele in enumerate(sum_x_axis_list):
- num = num % 2
- if ele == num:
- # print(i)
- num = num + 1
- split_x_index.append(index)
- # print('length: ', len(split_x_index), split_x_index)
- return split_x_index
- y_break_points_list = axix_break_point(img_arr, 1, axis=1)
- x_break_points_list = axix_break_point(img_arr, 1, axis=0)
- all_coordinates = []
- for i in range(0, len(y_break_points_list), 2): # y轴分组
- ymin = y_break_points_list[i]
- ymax = y_break_points_list[i + 1]
- for j in range(0, len(x_break_points_list), 2):
- xmin = x_break_points_list[j]
- xmax = x_break_points_list[j + 1]
- all_coordinates.append([xmin, ymin, xmax, ymax])
- return all_coordinates
- def get_choice_box_coordinate(word_result_list, choice_img, cv_box_list, choice_bbox_list):
- shape = choice_img.shape
- y, x = shape[0], shape[1]
- # cv2.imshow('ocr_region', ocr_region)
- # if cv2.waitKey(0) == 27:
- # cv2.destroyAllWindows()
- all_digital_list = []
- digital_model = re.compile(r'\d')
- for i, chars_dict in enumerate(word_result_list):
- chars_list = chars_dict['chars']
- for ele in chars_list:
- if digital_model.search(ele['char']):
- all_digital_list.append(ele)
- new_all_digital_list = []
- i = 1
- while i <= len(all_digital_list):
- pre_one = all_digital_list[i - 1]
- if i == len(all_digital_list):
- new_all_digital_list.append(pre_one)
- break
- rear_one = all_digital_list[i]
- condition1 = abs(pre_one['location']['top'] - rear_one['location']['top']) < pre_one['location'][
- 'height'] # 两字高度差小于一字高度
- condition2 = pre_one['location']['left'] + 2 * pre_one['location']['width'] > rear_one['location'][
- 'left'] # 某字宽度的2倍大于两字间间隔
- if condition1:
- if condition2:
- new_char = pre_one['char'] + rear_one['char']
- new_location = {'left': pre_one['location']['left'],
- 'top': min(pre_one['location']['top'], rear_one['location']['top']),
- 'width': rear_one['location']['left'] + rear_one['location']['width'] -
- pre_one['location']['left'],
- 'height': max(pre_one['location']['height'], rear_one['location']['height'])}
- new_all_digital_list.append({'char': new_char, 'location': new_location})
- i = i + 1 + 1
- else:
- new_all_digital_list.append(pre_one)
- i = i + 1
- else:
- new_all_digital_list.append(pre_one) # 遇到字符y轴相差过大就结束
- i = i + 1
- content_list = list()
- for index, box in enumerate(choice_bbox_list['regions']): # rcnn识别的框匹配题号
- box = box['bounding_box']
- box_coordinate = (box['xmin'], box['ymin'], box['xmax'], box['ymax'])
- horizontal = box['xmax'] - box['xmin'] >= box['ymax'] - box['ymin']
- vertical = box['xmax'] - box['xmin'] < box['ymax'] - box['ymin']
- choice_number = {'number': 99, 'location': box_coordinate}
- content_list.insert(index, choice_number)
- for digital in new_all_digital_list:
- digital_coordiante = (digital['location']['left'], digital['location']['top'],
- digital['location']['left'] + digital['location']['width'],
- digital['location']['top'] + digital['location']['height'])
- if utils.decide_coordinate_contains(digital_coordiante, box_coordinate):
- if horizontal:
- box['xmin'] = digital['location']['left'] + digital['location']['width'] + 1 # 从数字处截取
- if vertical:
- box['ymin'] = digital['location']['top'] + digital['location']['height'] + 1
- box_coordinate = (box['xmin'], box['ymin'], box['xmax'], box['ymax'])
- content_list[index]['number'] = digital['char']
- content_list[index]['location'] = box_coordinate
- break
- for box in content_list:
- box_coordinate = (box['location'][0], box['location'][1], box['location'][2], box['location'][3])
- mtx = []
- for cv_box in cv_box_list:
- if utils.decide_coordinate_contains(cv_box, box_coordinate): # 若fasterrcnn未识别到选项框,单独的ABCD也舍去
- mtx.append(cv_box)
- matrix = np.asarray(sorted(mtx))
- dif = matrix[1:, 0] - matrix[:-1, 2] # 后一个char的left与起一个char的right的差
- dif[dif < 0] = 0
- dif_length = np.mean(dif) # 小于平均间隔的合并
- block_list = utils.box_by_x_intervel(matrix, dif_length)
- # block_list = utils.box_by_x_intervel(matrix, 5)
- box['abcd'] = block_list
- return content_list
- def choice(left, top, image, choice_bbox_list, xml_path):
- a_z = '_ABCDEFGHIJKLMTUNOPQRSVWXYZ'
- t1 = time.time()
- word_result_list0 = get_ocr_text_and_coordinate(image, ocr_accuracy='accurate', language_type='ENG')
- t2 = time.time()
- print('choice ocr time cost: ', t2 - t1)
- # print(word_result_list0)
- # try:
- # intervel_x = get_interval(word_result_list0)
- # except Exception:
- # intervel_x = 15
- intervel_x = 3
- img = preprocess(image, intervel_x, 3)
- cv_box_list0 = box_coordinates(img)
- content_list = get_choice_box_coordinate(word_result_list0, image, cv_box_list0, choice_bbox_list)
- tree = ET.parse(xml_path) # xml tree
- w = content_list[0]['location'][2] - content_list[0]['location'][0]
- h = content_list[0]['location'][3] - content_list[0]['location'][1]
- def xml(xml_tree, sorted_abcd_list, bias=0):
- ii = 0
- for i, choice_bbox in enumerate(sorted_abcd_list):
- area = (choice_bbox[2] - choice_bbox[0]) * (choice_bbox[3] - choice_bbox[1])
- if area > 400:
- name = '{:02d}_{}'.format(int(choice['number']), a_z[ii + bias])
- xml_tree = utils.create_xml(name, xml_tree,
- choice_bbox[0] + left, choice_bbox[1] + top, choice_bbox[2] + left,
- choice_bbox[3] + top)
- ii += 1
- return xml_tree
- def get_json(ajson_list, sorted_abcd_list, bias=0):
- ii = 0
- for i, choice_bbox in enumerate(sorted_abcd_list):
- area = (choice_bbox[2] - choice_bbox[0]) * (choice_bbox[3] - choice_bbox[1])
- if area > 400:
- name = '{:02d}_{}'.format(int(choice['number']), a_z[ii + bias])
- region = [choice_bbox[0] + left, choice_bbox[1] + top, choice_bbox[2] + left, choice_bbox[3] + top]
- ajson_list.append({'number': name, 'region': region})
- ii += 1
- return ajson_list
- json_list = []
- for index_num, choice in enumerate(content_list):
- abcd = choice['abcd']
- if int(choice['number']) == 99:
- if w >= h:
- tree = xml(tree, sorted(abcd))
- json_list = get_json(json_list, sorted(abcd))
- else:
- tree = xml(tree, sorted(abcd, key=lambda x: (x[1], x[0])))
- json_list = get_json(json_list, sorted(abcd, key=lambda x: (x[1], x[0])))
- else:
- if w >= h:
- tree = xml(tree, sorted(abcd), bias=1)
- json_list = get_json(json_list, sorted(abcd), bias=1)
- else:
- tree = xml(tree, sorted(abcd, key=lambda x: (x[1], x[0])), bias=1)
- json_list = get_json(json_list, sorted(abcd, key=lambda x: (x[1], x[0])), bias=1)
- tree.write(xml_path)
- return json_list
- def get_number_by_enlarge_choice_m(image, choice_m_region_list, xml_path):
- a_z = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
- choice_m_dict_list = [] # choice_m region with same index
- choice_m_enlarge = []
- left, top, right, bottom = 9999, 9999, 0, 0
- for _, box in enumerate(choice_m_region_list):
- box = box['bounding_box']
- m_left, m_top = box['xmin'], box['ymin'],
- width, height = box['xmax'] - box['xmin'], box['ymax'] - box['ymin']
- box_coordinate = (m_left, m_top, box['xmax'], box['ymax'])
- single_choice_m = utils.crop_region_direct(image, box_coordinate)
- row_col_dict = get_choice_m_row_and_col(m_left, m_top, single_choice_m)
- choice_m_dict_list.append(row_col_dict)
- box_coordinate_enlarge = (
- m_left - int(width / 2), m_top - int(height / 2), box['xmax'], box['ymax']) # 扩大的choice_m, 多个分散choice_m
- choice_m_enlarge.append(box_coordinate_enlarge)
- left = min(left, box_coordinate_enlarge[0])
- top = min(top, box_coordinate_enlarge[1])
- right = max(right, box_coordinate_enlarge[2])
- bottom = max(bottom, box_coordinate_enlarge[3])
- choice_whole_region = utils.crop_region_direct(image, (left, top, right, bottom))
- # cv2.imwrite(r'C:\Users\Administrator\Desktop\test\sheet\choice_enlarge.jpg', choice_whole_region)
- # cv2.imshow('img', choice_whole_region)
- # cv2.waitKey(0)
- # cv2.destroyAllWindows()
- choice_region_text = get_ocr_text_and_coordinate(choice_whole_region)
- all_digital_list = []
- pattern = re.compile(r'\d')
- for i, chars_dict in enumerate(choice_region_text):
- chars_list = chars_dict['chars']
- for ele in chars_list:
- if pattern.search(ele['char']):
- all_digital_list.append(ele)
- combined_digital_list = utils.combine_char(all_digital_list)
- direction_list = []
- for index, enlarge_box in enumerate(choice_m_enlarge):
- digital_list = []
- xmin, ymin, xmax, ymax = 9999, 9999, 0, 0
- choice_m_dict = choice_m_dict_list[index]
- choice_m_dict_box = (choice_m_dict['bounding_box']['xmin'], choice_m_dict['bounding_box']['ymin'],
- choice_m_dict['bounding_box']['xmax'], choice_m_dict['bounding_box']['ymax'],)
- for jndex, digital_box in enumerate(combined_digital_list):
- digital_coordinate = (digital_box['location']['left'] + left,
- digital_box['location']['top'] + top,
- digital_box['location']['left'] + digital_box['location']['width'] + left,
- digital_box['location']['top'] + digital_box['location']['height'] + top)
- digital_box.update({'coordinate': digital_coordinate})
- if (utils.decide_coordinate_contains(digital_coordinate, enlarge_box)) and not \
- (utils.decide_coordinate_contains(digital_coordinate, choice_m_dict_box)):
- digital_list.append(digital_box)
- xmin = min(xmin, digital_box['coordinate'][0])
- ymin = min(ymin, digital_box['coordinate'][1])
- xmax = max(xmax, digital_box['coordinate'][2])
- ymax = max(ymax, digital_box['coordinate'][3])
- digital_list_coordinate = (xmin, ymin, xmax, ymax)
- direction = utils.decide_choice_m_left_top(digital_list_coordinate, choice_m_dict_box)
- if int(direction):
- choice_m_dict['direction'] = direction
- direction_list.append(direction)
- if direction == '180': # 数字垂直排列
- std_num_length = choice_m_dict['rows']
- choice_option = a_z[:choice_m_dict['cols']].replace('', ',')[1:-1]
- default_points = [-1] * std_num_length
- choice_m_dict.update({'option': choice_option, 'default_points': default_points})
- sorted(digital_list, key=lambda k: k.get('coordinate')[1])
- choice_ymin = choice_m_dict['bounding_box']['ymin']
- single_height = choice_m_dict['single_height']
- mean_interval = ((choice_m_dict['bounding_box']['ymax'] - choice_m_dict['bounding_box']['ymin'])
- - single_height * std_num_length) / (std_num_length - 1)
- spilt_index = [choice_ymin - mean_interval / 2 + (single_height + mean_interval) * ele for ele in
- range(std_num_length + 1)]
- number_list = [-1] * std_num_length
- number_location = [(-1, -1, -1, -1)] * std_num_length
- for i in range(0, len(spilt_index) - 1):
- start = spilt_index[i]
- end = spilt_index[i + 1]
- number_location[i] = (xmin, start, xmax, end)
- for digital_coordinate in digital_list:
- middle_y = (digital_coordinate['coordinate'][3] - digital_coordinate['coordinate'][1]) / 2 + \
- digital_coordinate['coordinate'][1]
- middle_x = (digital_coordinate['coordinate'][2] - digital_coordinate['coordinate'][0]) / 2 + \
- digital_coordinate['coordinate'][0]
- if (start <= middle_y <= end
- and
- middle_x < choice_m_dict['bounding_box']['xmin']): # 数字在choice_m外侧
- number_list[i] = int(digital_coordinate['char'])
- number_location[i] = digital_coordinate['coordinate']
- number_list = _infer_number(number_list)
- choice_m_dict['number'] = _infer_number(number_list)
- # choice_m_dict['number'] = [{'number': number,
- # 'location': {'xmin': xi, 'ymin': yi, 'xmax': xm, 'ymax': ym}}
- # for number in number_list
- # for (xi, yi, xm, ym) in number_location]
- if direction == '90': # 数字水平排列
- std_num_length = choice_m_dict['cols']
- choice_option = a_z[:std_num_length].replace('', ',')[1:-1]
- default_points = [-1] * std_num_length
- choice_m_dict.update({'option': choice_option, 'default_points': default_points})
- sorted(digital_list, key=lambda k: k.get('coordinate')[0])
- choice_xmin = choice_m_dict['bounding_box']['ymin']
- single_width = choice_m_dict['single_width']
- mean_interval = ((choice_m_dict['bounding_box']['xmax'] - choice_m_dict['bounding_box']['xmin'])
- - single_width * std_num_length) / (std_num_length - 1)
- spilt_index = [choice_xmin - mean_interval / 2 + (single_width + mean_interval) * ele for ele in
- range(std_num_length)]
- number_list = [-1] * std_num_length
- number_location = [(-1, -1, -1, -1)] * std_num_length
- for i in range(0, len(spilt_index) - 1):
- start = spilt_index[i]
- end = spilt_index[i + 1]
- number_location[i] = (start, ymin, end, ymax)
- for digital_coordinate in digital_list:
- middle_y = (digital_coordinate['coordinate'][3] - digital_coordinate['coordinate'][1]) / 2 + \
- digital_coordinate['coordinate'][1]
- middle_x = (digital_coordinate['coordinate'][2] - digital_coordinate['coordinate'][0]) / 2 + \
- digital_coordinate['coordinate'][0]
- if start <= middle_x <= end and middle_y < choice_m_dict['bounding_box']['ymin']:
- number_list[i] = int(digital_coordinate['char'])
- number_location[i] = digital_coordinate['coordinate']
- number_list = _infer_number(number_list)
- choice_m_dict['number'] = _infer_number(number_list)
- # choice_m_dict['number'] = [{'number': number,
- # 'location': {'xmin': xi, 'ymin': yi, 'xmax': xm, 'ymax': ym}}
- # for number in number_list
- # for (xi, yi, xm, ym) in number_location]
- else:
- choice_m_dict['direction'] = '0'
- choice_m_dict['number'] = [-1]
- choice_m_dict['default_points'] = [-1]
- count180 = ','.join(direction_list).count('180')
- count90 = ','.join(direction_list).count('90')
- infer_direction = ['180', '90'][[count180, count90].index(max(count180, count90))]
- for ele in choice_m_dict_list:
- if ele['direction'] != '0':
- ele.update({'direction': infer_direction})
- # tree = ET.parse(xml_path) # xml tree
- # for index_num, choice_box in enumerate(choice_m_dict_list):
- # if len(choice_box['bounding_box']) > 0:
- # abcd = choice_box['bounding_box']
- # number = str(choice_box['number'])
- # name = '{}_{}*{}_{}_{}'.format('choice_m', choice_box['rows'],
- # choice_box['cols'], choice_box['direction'],
- # number)
- # tree = utils.create_xml(name, tree,
- # abcd['xmin'], abcd['ymin'],
- # abcd['xmax'], abcd['ymax'])
- #
- # tree.write(xml_path)
- return choice_m_dict_list
- def _infer_number(number_list):
- if -1 not in number_list or sum(number_list) == -1 * len(number_list):
- return number_list
- else:
- for n_index in range(0, len(number_list) - 1):
- if n_index == 0:
- if number_list[n_index] != -1:
- if len(number_list) > 1 and number_list[n_index + 1] == -1:
- number_list[n_index + 1] = number_list[n_index] + 1
- if number_list[n_index] != -1:
- if number_list[n_index - 1] == -1:
- number_list[n_index - 1] = number_list[n_index] - 1
- if number_list[n_index + 1] == -1:
- number_list[n_index + 1] = number_list[n_index] + 1
- return _infer_number(number_list)
|